Learning Goals:

  1. Explain how social norms can control whether prejudice is expressed
  2. Explain the difference between normative and empirical expectations

Motivation:

Rather than change attitudes, maybe we can change behaviors.

People can still be prejudiced, but they may not act on or express those prejudices if they know that such behavior is counter-normative.

A Simple Example: (not) smoking in certain places


Elements of social norms

Social norms spell out what we should (or should not) do in certain situations.

In other words, they are beliefs / expectations about how others think we ought to behave.

Some people call these beliefs “injunctive norms” or simply “normative expectations”.

Examples:

As we have seen in a previous class (week 6), one consequence of the existence of a social norm is that norm violations will be sanctioned. People are therefore usually keen to avoid violating norms (at least when others can see them).

This might explain why election polls often under-estimate support for far-right political candidates – that is, people who actually plan to vote for the far-right on election day are embarassed to reveal their true voting intentions to pollsters.

Another Example of How Norms Can Constrain the Expression of Prejudice


(from Vicente Valentim)

In Spain, people vote using voting cards:


To preserve privacy, voters usually make their selection in private booths.

But in one municipality, 4 elections were held on the same day, and they ran out of voting booths! Thus, for one out of the 4 elections, voters had to pick up their cards from a public table (making their votes potentially observable by others).

When voting was potentially public, people were significantly less likely to support the far-right Partido Popular.


Normative compliance

Just because a social norm exists doesn’t guarantee that people will follow it, or that norm violations will be sanctioned.

Example: you know littering is wrong, but…

Normative compliance depends crucially on the belief that others are also following the norm.

Some people use the term “descriptive norms” or “empirical expectations” to refer to these beliefs about others’ behavior .


To summarize:

  1. Injunctive norms (or normative expecations) are our beliefs about what others think we should do
    • i.e. we believe that most people think that littering is wrong
  2. Descriptive norms (or empirical expectations) are our beliefs about what others actually do
    • i.e. we believe that most people do (not) litter

These two sets of beliefs need not coincide!

Consider: in many countries, most people think that corruption is wrong. Yet lots of people are corrupt anyways! Indeed, the fact that other people are violating the honesty norm gives me license / justification to also break the rules.


Norms against hate speech

Even internet trolls know that most people find hate speech socially unacceptable.

So there’s no point in trying to change anyone’s beliefs about injunctive norms.

This leaves us with two options for reducing hate speech:

  1. Sanction norm violations (e.g. counter-speaking)
  2. Changing beliefs about whether other people also engage in hate speech online (e.g. censoring)

Alvarez and Winter experiment

Message Board
Design
Censored Treatment
Counterspeaking Treatment
Data Collection
Creating a Hate Score
Results

Munger Experiment

Design


Four treatments:

  • White avatar, few followers
  • Black avatar, few followers
  • White avatar, many followers
  • Black avatar, many followers
Results



Normative Unravelling

A Case Study:

Even before his election to the US Presidency, Donald Trump was already infamous for his embrace of racist (and sexist) views. After all, he announced his candidacy by calling Mexicans “drug dealers” and “rapists.”

Polls leading up to the 2016 US Presidential election consistently predict a moderate lead for Hillary Clinton over Donald Trump. Very few people seriously thought that a majority of Americans would vote for him.

Of course, in a shocking turn of events, Trump beat Clinton on election day. This result surprised most Americans.

Divide into groups, and consider the following hypothetical experiment (based off a real design by Bursztyn et al.):

Group Assignment


Details


You recruit a random sample of Americans to take part in your study. You pay them 10 USD for taking your survey. After answering some basic demographic and political questions, you ask them whether they would like to donate their 10 USD participation fee to an organization called the Federation for American Immigration Reform (FAIR).

Survey participants read the following information about FAIR:

The Federation for American Immigration Reform is an immigration-reduction organization of concerned individuals who believe that immigration laws must be reformed and seeks to reduce overall immigration (both legal and illegal) into the United States. The founder of FAIR is John Tanton, author of “The Immigration Invasion” who wrote “I’ve come to the point of view that for European-American society and culture to persist requires a European-American majority, and a clear one at that.”

Participants’ decision to donate or not is your outcome variable.


Unbeknownst to participants, your survey is actually a randomized experiment.

Participants in the control group receive information that their donation decisions will be made in private:

Note: just like any other answer to this survey, also your donation decision will be completely anonymous. No one, not even the researchers, will be able to match your decision to your name.

But this assurance of privacy is not provided to participants in the treatment group. Instead, they are told:

Important: in order to ensure the quality of the data collected, a member of the research team might personally contact you to verify your answers.

Either the treatment or the control prompt is shown to participants immediate before they are asked to make their donation decision.

Imagine you were to field this study twice: once before the 2016 election, and once afterwards.

  1. What do you think would be the difference in donation rates between the treatment and control groups in the pre-election wave?
  2. Do you think this difference would become larger, become smaller or stay the same in the post-election wave?
  3. Explain your reasoning.

A Modified Study


You run a study which is similar to that just described, but with a few changes.

First, rather than sampling from the US population as a whole, you only sample political independents (i.e. people not affiliated with either the Democratic or Republican parties). For our purposes, suppose further that political independents are neither particularly “woke” nor particularly racist.

Second, there is only the treatment condition described above. You don’t have a control condition.

Third, you field the study as a 2-wave panel. That is, you are able to re-interview 100% of the same people in both waves, and thus compare how their decisions change over time.

Finally, you have a question in your pre-election survey which asks people whether they think Donald Trump will win the election. Amongst your sample of political independents, how people answer this question is uncorrelated with their racial attitudes.

  1. Let’s first consider people who think Donald Trump will lose the election. How, if at all, would their decisions change between the two waves?
  2. What about people who think that Trump will win the election? How, if at all, would their decisions change?